LLM Agent Skills
Skills developed by Carl Kibler to build better products and some random tasks.
Should work in Claude Code, Codex, and others.
Install
Claude Code
Add this repo as a Claude Code marketplace:
/plugin marketplace add carlkibler/agent-skills
Start with these two — you won't be disappointed:
/plugin install pre-mortem@carl-tools
/plugin install kindness-check@carl-tools
/pre-mortem:run — Point it at any project or idea. Parallel agents attack it from different failure angles and return a ranked, concrete list of what will actually kill the launch.
/kindness-check:run — Run it on any feature after you think it's done. Reviews through four lenses (user, machine, developer, support) and surfaces what "works" but shouldn't ship.
Or install anything from the full list below, or browse via /plugin → Discover tab.
After installing, run /reload-plugins to activate.
Codex CLI / Codex app
This repo now includes native Codex integration in two forms:
- repo-local skill discovery via
.agents/skills/
- repo-local Codex marketplace via
.agents/plugins/marketplace.json and plugins/*/.codex-plugin/plugin.json
To use the skills directly from a checked-out repo, open the repo in Codex and type $ or /skills. Codex will discover the skills from .agents/skills automatically.
To test the packaged Codex plugins locally, restart Codex after cloning the repo, open the plugin directory, and look for the Carl Tools local marketplace sourced from this repo.
Codex web / cloud
Repo-local skills under .agents/skills/ travel with the repository and are available when Codex checks out the repo in a cloud task.
A few skills are intentionally local-machine-centric and work best in CLI/app sessions on your own machine, not generic cloud containers. Examples: chezmoi-drift, profile-me, and parts of handle-pr.
Cloud-friendlier skills: pre-mortem, trust-audit, support-storm, first-contact, parallel-isolated-app-testing, second-opinions, and wifi-qr.
Usage
Skills are namespaced by plugin name. Invoke directly or let Claude trigger them automatically:
/pre-mortem:run
/kindness-check:run
Or just ask naturally — Claude will invoke the right skill based on context.
Skills
Better Products
Find failure modes, trust problems, and support burden before your users do.
| Skill | |
|---|
| first-contact | Red-team onboarding and first-run experience for abandonment traps
/plugin install first-contact@carl-tools |
| launch-sequence | Full pre-launch gauntlet with GO/CAUTION/NO-GO verdict
/plugin install launch-sequence@carl-tools |
| post-mortem | Blameless post-mortem that feeds learnings back into your skill collection
/plugin install post-mortem@carl-tools |
| pre-mortem | Multi-agent project pre-mortem — ranked risks with mitigations
/plugin install pre-mortem@carl-tools |
| scope-hammer | Compress brainstorm output into MVP scope with DELETE/MOCK/ALREADY EXISTS/SHIP classification
/plugin install scope-hammer@carl-tools |
| status-copy-trust-audit | Make status/update/doctor output explain exactly what changed and why
/plugin install status-copy-trust-audit@carl-tools |
| support-storm | Simulate the support emails and refunds a launch will generate
/plugin install support-storm@carl-tools |
| trust-audit | Audit a product's trust surface: permissions, privacy, billing, and silent failures
/plugin install trust-audit@carl-tools |
| visual-qa-loop | Repeatable visual screenshots and review notes for UI changes
/plugin install visual-qa-loop@carl-tools |
| wide-open-brainstorm | Multi-model brainstorming room for product strategy — serious plus whimsical, multi-altitude ideation
/plugin install wide-open-brainstorm@carl-tools |
Dev Workflow
Tools for the day-to-day of writing and reviewing code.
